Google Releases Three Embodied AI Models, Solving Robot 'Temporal Intelligence'
aigclink · x · 2026-07-31
Google released three new models for embodied AI, focusing on the challenge of 'temporal intelligence'—knowing when a task is fully completed:
- Gemini Robotics 2: A closed-source VLA model acting as the 'body', controlling humanoid robots from feet to fingertips.
- Gemini Robotics ER 2: An embodied reasoning model acting as the 'brain'. It breaks through temporal intelligence by identifying key action frames with 91.3% accuracy (0.96s avg error), enabling smooth 'thinking while doing', and supporting multi-robot collaboration (closed-source).
- On-Device 2: An open-source locally-run model that can adapt to a completely new robot body in just hours.
This architecture decouples the reasoning brain from the executing body, shifting the robotics competition from mere mobility to precision and awareness.
Related event: Google DeepMind Unveils Gemini Robotics 2(69 posts)→
More from Embodied
- AI Makes Call of Duty Heartbeat Sensor Real: Sensing the Physical World via RF — bilawalsidhu · 2026-08-01
- VLM run Integrates Gemini Robotics ER 2 into Orion 2 Visual Agent Harness — spillai · 2026-08-01
- Robot Goal-Reaching Test: Model Unexpectedly Breaks Into a Funny Dance — carlosdponx · 2026-08-01
- SAMRA Home Companion Robot: Multi-Axis Design Tailored for Children — tobowers · 2026-08-01
- Musk Calls Human Language 'Lossy Compression', Predicts Neuralink Will Replace Speech — r0ck3t23 · 2026-08-01
- Google DeepMind Unveils Gemini Robotics 2 Model — TheAIGRID · 2026-08-01